SWOV Catalogus

341407

Using naturalistic driving data to assess vehicle-to-vehicle crashes involving fleet drivers.
20150874 ST [electronic version only]
Carney, C. McGehee, D. Harland, K. Weiss, M. & Raby, M.
Washington, D.C., American Automobile Association AAA Foundation for Traffic Safety, 2015, 18 p., 9 ref.

Samenvatting Objective, detailed and accurate data is critically important in the effort to determine the causes and contributing factors of crashes. In the past, the only way to obtain such information for a large number of crashes was to use data collected from police reports. While information gathered this way is helpful, it has many limitations. More recently, in-vehicle event recorders (IVERs) have become a widely accepted means of gathering crash data, both in research and real-world applications. In this study, we conducted the first-ever large-scale examination of naturalistic crash data. Other naturalistic studies have investigated only a small number of crashes or used near crashes as a proxy for real crashes. In contrast, this project examined hundreds of actual crashes from a naturalistic driving database. The data allowed us to examine behaviours and potential contributing factors in the seconds leading up to the collision, and provided information not available in police reports. A coding scheme was developed specifically for this study, and video data were coded with the goal of identifying the factors that contributed to crashes–in particular the prevalence of potentially distracting driver behaviours and drowsiness. The study addressed the following research questions: • What were the roadway and environmental conditions at the time of the crash? • What were the critical events and potential contributing factors leading up to the crash and did these differ by crash type? • What driver behaviours were present in the vehicle prior to the crash and did these differ by crash type? • How did driver response times and eyes-off-road time differ relative to certain driver behaviours and crash types? • Could drowsy driving be detected using this type of crash data? Understanding the prevalence of factors that potentially contribute to crashes will provide a significant societal benefit and advance the field of traffic and crash safety. More specifically, information regarding what is happening inside the vehicle during the seconds before a crash can be used to pinpoint automotive safety systems and technologies that might best mitigate certain types of crashes. Lytx, a company that has been collecting data using in-vehicle event recorders (IVERS) for over a decade, provided the crash data. The DriveCam system collects video, audio and accelerometer data when a driver triggers the device by hard braking, fast cornering, or an impact that exceeds a certain g-force. Each video is 12 s long, and provides information on the 8 s before and 4 s after the trigger. The system has a wide range of applications– families use them to help young drivers as they begin to drive independently, while over 950 commercial and government fleets employ them for fleet management. As part of this study, 777 crashes from the fleet driver database were made available for review. In order to eliminate minor curb strikes from the analysis, those crashes in which the vehicle sustained impacts of less than 1g were excluded. Crashes in which the DriveCam-equipped vehicle was struck from behind were also eliminated from this particular analysis. Additional videos were excluded for other reasons (e.g., animal strikes, video problems, or the driver not being a member of the fleet). Consequently, nearly 250 moderate-to-severe vehicle-to-vehicle crashes remained for analysis in the current study. A coding methodology focused on identifying the factors present in the seconds leading up to the crashes was developed specifically for gathering information from the videos. Development of the coding method began with a thorough review of existing crash coding from government, academic and industry sources. In all, 64 data elements were identified as relevant to the project goals. These were narrowed down according to their relevance to the project and ability to be coded reliably. In the end, 24 data elements were selected for inclusion in the coding methodology. These elements were specific to environmental conditions, contributing circumstances (e.g., inadequate surveillance, running traffic signals), and driver and passenger behaviours. Some of the elements could have multiple data coded (i.e., multiple driver behaviours occurred within one crash segment). Each crash, in particular the 6 s leading up to the crash (this time frame was selected to ensure results were comparable to other naturalistic driving studies), was double coded by two University of Iowa (UI) analysts and mediated by a third when necessary. For this study, we analysed 247 moderate-to-severe vehicle-to-vehicle crashes in which the force of the impact was 1.0g or greater. While the extent of any injuries sustained in the crashes was not evident from the videos, it is known that no fatal crashes were included in the analysis. However, it is likely that most, if not all, of the crashes would have resulted in a police report being filed. The majority of fleet drivers were between the ages of 30 and 64 (73.3%) and nearly 87% were male. Of the crashes coded, the majority were angle (52%) and rear-end (41%) crashes. Approximately 10% of crashes were due to environmental factors, such as poor road conditions. Crashes did not seem more likely to occur on any particular day of the workweek or time of day. The critical pre-crash event in 97% of rear-end crashes was another vehicle in the driver’s lane decelerating or stopping in the roadway. As mentioned above, rear-end crashes in which the DriveCam-equipped vehicle was struck from behind were omitted from the analysis. As to angle crashes, the participant’s vehicle crossed the centerline or was turning at an intersection in 45% of crashes, while another vehicle encroaching accounted for 51%. Regardless of fault, in 84% of crashes, the driver contributed to the crash in some way. Recognition errors, such as inadequate surveillance and engaging in a potentially distracting behaviour were observed in 71% of crashes. Decision errors, such as following too closely and running stop signs and lights, were coded in 40% of crashes. Performance errors, such as losing control of the vehicle, occurred in only 3% of crashes. Attending to a location either outside or inside the vehicle that was not relevant to safe operation of the vehicle were the two most frequently coded driver behaviours seen in the six seconds leading up to a crash. These behaviours were associated with recognition errors such as inattentive/engaged in extraneous behaviours and inadequate surveillance. Cell phone use was the third most frequent driver behaviour observed, occurring in 8.3% of crashes. Among crashes with driver cell phone use, the driver was coded as operating/looking at a phone during 53% of these events, talking/listening in 31%, and cell phone use was coded as likely but not visible in 26%. Ninety-six percent of all cell phone-related behaviours happened when the driver was alone in the vehicle. Operating or looking at the phone never occurred when there were passengers present. When a driver was alone, he/she was seen engaging in potentially distracting behaviours in slightly more than half of crashes (52%). Cell phone use was 3.5 times as likely, and personal grooming and talking to oneself were almost twice as likely when the driver was alone. When passengers were present, having a conversation with the driver was the most common behaviour observed, occurring in 21% of crashes. Drivers involved in a rear-end crash were nearly twice as likely to be seen engaging in non-driving-related activities during the six seconds leading up to the crash compared to drivers of angle crashes. In addition, the average time the driver’s eyes were off of the forward roadway was more than 4 times as long for rear-end crashes than for angle crashes (3.2 vs. 0.7s). There are limitations associated with event-triggered driving data that make detecting drowsy driving extremely difficult. For this study, only four of the 229 fleet crashes examined contained evidence of drowsy driving. Use of IVERs in naturalistic driving allows researchers a unique view into the vehicle, and provides invaluable information regarding the behavioural and environmental factors present before a crash. The data gathered offers a much more detailed context relative to police reports and other crash databases, and allows more micro-level analyses to be conducted. This study examines the roadway and environmental conditions present in different types of crashes. It describes the critical events and contributing factors that lead up to crashes, and how they vary by crash type. It also provides information regarding the effect certain driver behaviours have on reaction time and eyes-off-road time. Finally, it is the first naturalistic study of moderate-to-severe crashes to examine driver and passenger behaviours for a variety of crash types among fleet drivers. The results of this study indicate that there are different driver behaviours and contributing circumstances present for rear-end vs. angle crashes. The most common driver behaviour seen was inadequate surveillance, with attending inside or outside the vehicle to an unknown location being coded most often. However, fleet drivers were more likely to be seen engaging in these potentially distracting behaviours when they were alone in the vehicle. Additionally, drivers involved in a rear-end crash were more likely to engage in a potentially distracting behaviour and had total eyes-off-road times that were four times as long as than those involved in angle crashes. (Author/publisher)
Full-text
Dossier
Suggestie? Neem contact op met de SWOV bibliotheek voor uw opmerkingen
Copyright © SWOV | Juridisch voorbehoud | Contact